Does AI and Human Advice Mitigate Punishment for Selfish Behavior? An Experiment on AI ethics From a Psychological Perspective

People don't blame AI for selfish decisions the way they blame humans: While decision-makers who followed AI advice were perceived as more responsible for their actions than those following human advice, they received similar levels of punishment. This suggests people may hold AI less accountable as an advisor, even though they still hold the person making the decision responsible.

Following bad advice makes selfish behavior seem worse, not better: When decision-makers acted selfishly after receiving prosocial advice, they were punished more harshly than those who acted selfishly with no advice at all. This indicates that ignoring good advice compounds moral judgment, regardless of whether the advice came from AI or humans.

Selfish advice does provide some legal/social protection: Selfish behavior was punished more leniently when preceded by selfish advice compared to no advice. This raises concerns that people relying on AI that encourages self-interested behavior may face fewer social or reputational consequences for acting on it.

What AI recommends matters more than that it's AI: The content of the advice—whether it promoted selfish or prosocial behavior—was the primary factor shaping how harshly people were punished. The fact that advice came from an algorithm versus a human had minimal impact on punishment levels, suggesting people evaluate the advice itself rather than its source.

Responsibility and punishment don't always align: Decision-makers were seen as more responsible when following AI advice, yet weren't punished differently, highlighting a gap between how we assign blame and how we actually hold people accountable.